Mind your step
Why developers of AI need more than a guardrail
If there’s one thing I have noticed from the hundreds of articles on why AI is good/bad (delete as appropriate), it’s that the word ‘guardrail’ is not only trending but is doing a lot of heavy lifting when deployed.
The word may be more commonly used in management theory, but the ‘guardrail’ is now increasingly cited as a necessary measure to guide the development of AI. From the floor of Congress to the pages of the Financial Times, the great and good argue that putting a guardrail in place will ensure that AI is developed in a way that doesn’t spell disaster for the human race.
In which case, I would suggest the word is wholly inadequate, and I have some better suggestions instead.
We are told that AI may take our jobs — so perhaps those that develop it need to wear safety shoes instead, to ensure they tread carefully. (There will be an awful lot of steel-toed safety footwear going around when AI has removed every human operative from factory floors.)
But it’s not just manual jobs that are at risk. According to Goldman Sachs, architects and engineers will also be replaced by AI. So maybe hard hats should be worn by our valiant AI developers?
And, as numerous breathless commentators tell us, AI-enabled robot cars will one day rule the roads. It seems ironic that we might revert to wearing crash helmets while reclining in the plush back seats of self-driving cars, but this type of protection might be more effective than a mere guardrail.
Forget your jobs, others warn; AI is coming for democracy itself. Historian Yuval Harari has argued that, as AI gains command of our language, it could destroy our ability to have the very conversations necessary to maintain democracy. Maybe we need to employ gags, then, instead of guardrails, and make it a little harder for AI to speak for us?
Others are even more scared of a future with AI. No less than the Godfather of AI, ex-Googler Geoffrey Hinton has warned of the potential destruction of humankind if AI is given control of the numerous systems that we rely on.
I can’t see that a guardrail will really cut it, then. A circuit breaker, kill switch or a fire axe with enough heft to sever a data cable might, but I fear it will be too late to break the glass in an emergency and reach for the fusebox once AI has figured out the launch codes and trained the world’s nuclear arsenals on us.
A guardrail helps tourists navigate the path along a deep ravine or makes the descent of stairs a little easier. It provides something to lean on if your cruise ship cabin faces the sea, and a version of the same offers something to grab onto when getting in and out of a bath.
Crucially, the presence of a guardrail doesn’t stop people from tripping, slipping or tumbling to their death, whether trekking in the Alps, sailing the seven seas, or indulging in candlelit bathing me-time at home. At least two people fall from a cruise liner every month, for heaven’s sake.
The guardrail is effectively useless; more a visual reminder of the need for care and attention than an adequate means of protection. It might only be made of rope or flimsy plastic, after all. Often, a guardrail solely exists to reduce corporate liability in the unfortunate event of an accident and a claim for damages. It’s a symbol, then; a signifier, a sop to health and safety regulations. It has as much ‘seriousness’ of purpose as the hard hats and hi-viz jackets that politicians favour for photo calls on brand-new factory floors.
On this basis, while the debate around the relative dangers of AI will continue ad infinitum, I suggest we agree at this point that a guardrail is probably a little underpowered for the fight we have lined up for it.
It’s hard to see how a guardrail can provide suitable protection from even the more benign effects of AI, such as spreading misinformation. At present, regulators can’t stop a TV station from lying about who won an election, and that’s before Fox TV has deployed artificial intelligence (although there is precious little sign of intelligence, artificial or otherwise, in that particular newsroom.)
Abandon hope all those who reach for the rope, for that guardrail will be useless when AI looks into the ravine below us, and push comes to shove. If Covid didn’t persuade you of the need to build a shelter in the garden, the empty reassurances of those developing AI with ‘guardrails’ in place just might. Make sure that shelter is off-grid, though, yeah?
